JPS6011359B2 - speech synthesizer - Google Patents

Description

DETAILED DESCRIPTION OF THE INVENTION The present invention relates to an analysis-synthesis type speech synthesizer capable of synthesizing and outputting not only speech but also musical tones such as distorted squares and sine waveforms.

Most of the speech synthesizers currently in practical use are based on the partial autocorrelation analysis synthesis method, and the circuit for performing synthesis calculations has come to be integrated on a single silicon chip.

Such a speech synthesizer generally integrates each functional circuit 100 on the synthesis side of the analysis and synthesis system shown in FIG. In the figure, 300 is a parameter file, which is means for storing the characteristic parameters of the voice analyzed and extracted by the analyzer 200, such as a read-only memory. The main part of this speech synthesizer generally has a circuit configuration as shown in the block diagram of FIG. 2.・Voiceless judgment code, amplitude, partial autocorrelation coefficient (
a decoder 110,1 that decodes the so-called K parameter)
20, 130, memories 111, 121, 131 for temporarily storing respective decoded parameters, memory 111
A pulse generation circuit 112 generates a pulse train corresponding to the value of the pitch parameter which is the output of a pulse generator 112, a white noise generation circuit 113 generates white noise to be used as an unvoiced sound resource, a voiced/unvoiced determination code 'This corresponds to a sound source signal. A sound source filter selection circuit 114 selects whether the pulse train is a white noise signal, an amplitude multiplication circuit 140 multiplies the sound source signal by the contents of the amplitude value memory 121, and a filter coefficient corresponding to the contents of the K parameter memory 131 is used to convert the sound source signal Digital filter 15 for extracting a predetermined frequency vector component from U
0, digital wave height value Y of digital filter 150,
It is composed of a D/A converter 160 that converts the signal into an analog signal y.

Of course, they are not shown in the diagram, but in addition to these, there are also a timing signal generation circuit and a decoder 1 necessary to operate each of these functional circuits with temporal timing.
10, 120, and 130, an interface circuit for sequentially importing time-series data obtained by speech analysis stored in an external memory, etc. are added to constitute a speech synthesizer. In such speech synthesizers, information compression is performed on analysis data in order to save memory for storing speech data, and even if one second of speech is compressed to about 2000 bits, the intelligibility is not significantly impaired. However, it can be put to practical use.

There are various compression methods, but as an example, the amplitude parameter is 4.
~6 bits, the pitch parameter is 5 to 6 bits, and the K parameter is called non-uniform bit distribution.
o in order of 5, 5, 4, 4, 4, 4, 4, 3, 3, 3 bits or 7, 5, 4, 4, 4, 3, 3, 3, 3, 3
assigned to a bit. Decoder 110 in FIG.
, 120, and 130 are for decoding these ink-coded parameter codes into purchase prices of analysis data, and form a table with the number of words corresponding to the number of bits.

Usually, due to circuit configuration constraints, the decoded digital value has an accuracy of about 10 bits. Further, each value of the decoding table is set by linear quantization between the upper limit value and the lower limit value of the analysis value, or by linear quantization after inverse hyperbolic function transformation. When the above-mentioned speech synthesizer synthesizes speech, it is possible to obtain synthesized speech with a considerably high degree of naturalness using a speech data memory with a capacity of 4.

However, for musical tones such as sine waves, sufficient sound quality cannot be obtained due to spectral distortion caused by quantization and modulation noise due to mismatch between the sound source frequency and the polar frequency of the digital filter 150. As detailed later,
It was impossible to construct musical scales using pure tones such as sine waves, or to generate musical tones with a fundamental frequency of several hundred Hz or more. The digital filter 150 is a multi-stage lattice type filter shown in FIG.
It consists of 3. This invention improves the above-mentioned speech synthesizer so that it can compose not only speech but also musical tones such as sine waves.

The principle of this invention will be explained below. When the number of poles is 1, the transfer function of an all-pole digital filter is Z (Z) = A/(
10QZ-10QZ-2)...{1}Znie-p-j2
medium fT [p: attenuation constant Qi: linear prediction coefficient f: frequency T: sampling period].

In the above formula, if the polar frequency is fr, the denominator of formula '11 =
From the simultaneous equations set as 0, Q,: -pcos2mR
T...■Q2=1e-2p holds true.

On the other hand, this filter evening's in/gurus response is an illusion = Ae
-friT'” in plsin2 (represented by 31).

{Equation 3' means a damped vibration waveform, and is a waveform suitable for musical tones. Next, the linear prediction coefficient Qi is related to the K parameter of the partial autocorrelation coefficient by the following equation through mathematical conversion processing. K,=Q,/(1-2)
………■K2=Q2 Therefore hni(1/2mT) cos-1 [(10e-2p)K,
/(friend-p)]ni(1/2T)cos-,[(1-K
2) K,ノ (2ノ-K2)]
・・・・・・・・・【5}p=-(1/2)ln(
1K2).

According to the formula {5}, the frequency of the damped vibration waveform is uniquely determined by the values of the K2 parameters, and the damping constant is uniquely determined by the K2 parameter. In the same equation, when K2 is in the range of -0.95 to -1.0, the extent to which a change in K2 affects the polar frequency is 1% or less, and there is no audible sense of pitch deviation. In this case, n in equation {5) is approximately given by the following equation, and h corresponds only to K. h≠(1/2mT
) cos-IKr…・'6} The above range of K2 values corresponds to the damping constant of 0 to 0.0256, i.e. 1/“ at approximately 40 sampling periods from a steady sinusoidal waveform with no damping.
This corresponds to a waveform that attenuates.

This is close to the attenuation characteristic of the sound of a natural musical instrument such as a piano instrument, and is suitable for musical sounds. On the other hand, the calculation algorithm of a one-stage digital filter configured for audio use is the sequential calculation formula shown in Table 1. Table 1 In this equation, Yi and bi are the intermediate values at the i stage of the forward wave and backward wave in the lattice filter, respectively, and (i
), i is the sampling number.

The filter output is q(i). The first sequential calculation formula is K
When 3 to K, o = 0, it functions as a one-pole digital filter, and when expressed using linear prediction coefficients Q, , Q2, considering equation 4', -, Tobaga n
This is equivalent to the expression -2...'7}.

Here, xn is a waveform value corresponding to the n-th sample period, xn-, xn-2 are values at the sample time one and two samples before xn, respectively, and U is a sound source signal value.
Determined by the transfer function of equation {1) - the impulse response control equation of the digital filter.In equation [7}, the sound source signal value U
It corresponds to xn when is the impulse.

Based on the above principle, K and K2 parameters are set to K
, = cos2 ge frT, K2 = 1e-2p, store these values in the memory of the decoder in advance, and drive the digital filter with an impulse to obtain a damped oscillation waveform. However, when the speech synthesizer according to the present invention uses a lattice-type digital filter for audio as a subordinate, if the calculation precision of the filter and the precision of the decoded values of the parameters are not sufficient, the theoretical value There was a problem that a normal damped vibration waveform could not be obtained. That is,
The multiplier precision of the lattice-type digital filter used is about 14 bits, and the precision of the decoded value is about 10 bits. It is known that only vibration waveforms can be obtained. One of the biggest causes of this is the accumulation of rounding errors in digital calculations, and another is the maximum value of the decoded value of the K2 parameter (the theoretically possible minimum value is -1.0, in this case, p=
0, which is a steady sinusoidal waveform), depending on the accuracy -
The problem is that it becomes larger than 1.0. For example 1
For 0-bit precision, the minimum value of K2 is approximately 10.998
The decay time in this case is about 0.125 seconds at the highest sampling frequency. The present invention aims to overcome the above-mentioned problems of the prior art and to obtain a steady sine waveform or a damped oscillation waveform with a long decay time without increasing the scale of the speech synthesizer.

FIG. 4 shows the digital filter 1 of the speech synthesizer of this invention.
500 examples are shown.

In the figure, 154 sets at least one or more lower bits to "1" or "0" depending on whether the multiplication result is positive or negative.
”Increase circuit that increases the absolute value by increasing the absolute value, 15
Reference numeral 5 indicates a musical tone/voice identification signal input terminal, and 156 indicates an identification signal inputted to the musical tone/voice identification signal input terminal 155, and when the voice is used, the output of the multiplier 152 is directly added to the adder 151.
This is a switching circuit that inputs the value obtained by increasing the output of the multiplier 152 by the increase circuit 154 to the adder 151 during musical tone. A partial circuit diagram of a specific embodiment of the increase circuit 154 and the switching circuit 156 is shown in FIG. In the same figure,
151 and 152 are an adder/subtractor and a multiplier, respectively, shown in FIG. 4, and the respective operation results are 14-bit two's complement fixed-point numbers. D,~D,4 is the multiplier 152
The multiplication result is K2×Q(i-1), where D, is the least significant bit, and D,4 is the most significant bit, which indicates the sign bit. 15
8 and 159 are logic gates, and 160 and 161 are inverters.

At the time of musical tone synthesis, the musical tone/voice identification signal is ``1'', and the outputs of the logic gates 158 and 159 have sign bits D and 4.
An inverted signal is output.

If the sign is “0” when the sign is positive and “1” when the sign is negative, the lower two bits of the multiplication result are “1” when the sign is positive and “1” when the sign is negative.
0'' is output from the gates 158 and 159. Therefore, the absolute value of K2×b2(i-1) is, on average,
2-812-12) will increase. That is, in the conventional digital filter 150, the adder 1
The value input to 5.1 is K2×b2(i-1), but in the configuration of this invention, on average
-1) Ten arts (2-.32-12), which means that the absolute value of K2 is equivalently increased, and it is possible to obtain an attenuation waveform with smaller attenuation. On the other hand, during voice synthesis, the musical tone and voice discrimination signals are "0", and the outputs of the logic gates 157 and 158 are ○, D2 and K, respectively.
The value of xQ(i-1) is input to the adder 151 without being incremented, and the problem of divergence occurring in the calculation process of the digital filter 1500 when the increase circuit 154 is used is eliminated. As described above, the speech synthesizer of the present invention can obtain high-quality synthesized waveforms for both speech and musical tones without substantially increasing the circuit scale of conventional speech synthesizers.

[Brief explanation of drawings]

Figure 1 is a block diagram of a conventional partial autocorrelation analysis and synthesis system for speech analysis and synthesis, Figure 2 is a block diagram of a secondary speech synthesizer, and Figure 3 is a circuit configuration of a secondary lattice-type multistage digital filter. 4 is a functional explanatory diagram of an embodiment of the digital filter used in the speech synthesizer of the present invention, and FIG. 5 is a partial circuit diagram of a specific example of the digital filter used in the speech synthesizer of the present invention. . 1 1 1, 121, 131... Memory means,
1 12...Pulse generator, 113...
・White noise generator, 151...Adder, 152・
... Multiplier, 153 ... Delay device, 154
...Increase circuit, 300...parameter file,
1500...Digital filter. Note that the same reference numerals in the figures indicate the same or corresponding parts. Figure 1 Figure 2 Figure 3 Figure 4 Figure 5

Claims (1)

[Claims] 1.Basically, it consists of a digital sound source signal generation circuit, a lattice-type multistage digital filter that is composed of an adder/subtractor, a delay device, and a multiplier and extracts a predetermined frequency spectrum component from the sound source signal, and a memory means that stores the coefficients of the digital filter. In the speech synthesizer of the partial autocorrelation analysis synthesis method, at least one lower bit or more of the multiplication result of the coefficient K_2 parameter of the last stage before the last stage of the lattice type multistage digital filter is An increase circuit is installed that slightly increases the absolute value of the multiplication result by setting it to “1” if it is negative, and “0” if it is negative, and synthesizing a constantly continuing sine waveform or a damped oscillation waveform with a long decay time. A speech synthesizer characterized by outputting.